Mistral Large 4 trails DeepSeek V4.1 Flash on agentic coding benchmarks

An independent test compared Mistral Large 4 (1.05T parameters, 52B active, 1M-token context) with DeepSeek V4.1 Flash on the Gauntlet and Forge 2.0 benchmarks. Mistral scored 0.5462 vs 0.6957 on Gauntlet and 0.1086/0.3304 vs 0.2422/0.6618 on Forge, while spending at least $21.53 across six runs versus about $0.62 for DeepSeek.

• Gauntlet: Mistral 0.5462 in 144 min for $4.93, DeepSeek 0.6957 in 23 min for $0.21
• Forge 2.0: 0.1086 and 0.3304 for Mistral vs 0.2422 and 0.6618 for DeepSeek
• Part of Mistral's Gauntlet loss came from a scorer bug, not the model
• Mistral Large 4 launched 6 October in Public Preview at half list price for two weeks